Multilingual Search using

Query Translation and

Collection Selection

Jacques Savoy, Pierre-Yves Berger

University of Neuchatel, Switzerland

www.unine.ch/info/clef/

Our IR Strategy

1. Effective monolingual IR system1. Effective monolingual IR system

(combining indexing schemes?)(combining indexing schemes?)

2. Combining query translation tools2. Combining query translation tools

3. Effective collection selection and 3. Effective collection selection and

merging strategiesmerging strategies

Effective Monolingual IR

1.1. Define a stopword listDefine a stopword list

2.2. Have a « good » stemmerHave a « good » stemmer

Removing only the feminineRemoving only the feminineand plural suffixes (inflections)and plural suffixes (inflections)Focus on Focus on nounsnouns and and adjectivesadjectives see www.unine.ch/info/clef/see www.unine.ch/info/clef/

Indexing Units

For Indo-European languages : Set of significant words

CJK: bigramswith stoplistwithout stemming

For Finnish, we used 4-grams.

IR Models

Probabilistic Okapi Prosit or

deviation from randomness

Vector-space Lnu-ltc tf-idf (ntc-ntc) binary (bnn-bnn)

Monolingual Evaluation

Model English French Portug.

Q=TD word word word

Okapi 0.5422 0.4685 0.4835

Prosit 0.5313 0.4568 0.4695

Lnu-ltc 0.4979 0.4349 0.4579

tf-idf 0.3706 0.3309 0.3708

binary 0.3005 0.2017 0.1834

Monolingual Evaluation

Model Finnish Russian

Q=TD word word

Okapi 0.4773 0.3800

Prosit 0.4620 0.3448

Lnu-ltc 0.4643 0.3794

tf-idf 0.3862 0.2716

binary 0.1859 0.1512

Monolingual Evaluation

Model Finnish Russian

Q=TD word 4-gram word 4-gram

Okapi 0.4773 0.5386 0.3800 0.2890

Prosit 0.4620 0.5357 0.3448 0.2879

Lnu-ltc 0.4643 0.5022 0.3794 0.2852

tf-idf 0.3862 0.4466 0.2716 0.1916

binary 0.1859 0.2387 0.1512 0.0373

Problems with Finnish

Finnish can be characterized by:

1. A large number of cases (12+),

2. Agglutinative (saunastani)

& compound construction (rakkauskirje)

3. Variations in the stem

(matto, maton, mattoja, …).

Do we need a deeper morpholgical analysis?

Improvement using…

Relevance feedback (Rocchio)?

Monolingual Evaluation

Q=TD English French Finnish Russian

Okapi 0.5422 0.4685 0.5386 0.3800

+PRF 0.5704 0.4851 0.5308 0.3945

+5.2% +3.5% -1.5% +3.8%

Prosit 0.5313 0.4568 0.5357 0.3448

+PRF 0.5742 0.4643 0.5684 0.3736

+8.1% +1.6% +6.1% +8.4%

Relevance FeedbackMAP after blind query-expansion

0 10 15 20 30 40 50 60 75 100# of terms added (from 3 best-ranked doc.)

MAP

0.36

0.38

0.4

0.42

0.44

0.46

0.48

0.5

Prosit Okapidtu-dtn Lnu-ltc

Improvement using…

Data fusion approaches (see the paper)

Improvement with the English, Finnish and Portuguese corpora

Hurt the MAP with the French and Russian collections

Bilingual IR

1. Machine-readable dictionaries (MRD)e.g., Babylon

2. Machine translation services (MT)e.g., FreeTranslation BabelFish

3. Parallel and/or comparable corpora(not used in this evaluation campaign)

Bilingual IR

“Tour de France WinnerWho won the Tour de France in 1995?”

(Babylon1) “France, vainqueurOrganisation Mondiale de la Santé, le, France,* 1995”

(Right) “Le vainqueur du Tour de FranceQui a gagné le tour de France en 1995 ?”

Bilingual EN->FR/FI/RU/PT

TDOkapi

Frenchword

Finnish4-gram

Russianword

Portugword

Manual 0.4685 0.5386 0.3800 0.4835

Baby1 0.3706 0.1965 0.2209 0.3071

FreeTra 0.3845 N/A 0.3067 0.4057

InterTra 0.2664 0.2653 0.1216 0.3277

Reverso 0.3830 N/A 0.2960 N/A

Combi. 0.4066 0.3042 0.3888 0.4204

Bilingual EN->FR/FI/RU/PT

TDOkapi

Frenchword

Finnish4-gram

Russianword

Portugword

Manual 0.4685 0.5386 0.3800 0.4835

Baby1 79.1% 36.5% 58.1% 63.5%

FreeTra 82.1% N/A 80.7% 83.9%

InterTra 56.9% 49.3% 32.0% 67.8%

Reverso 81.8% N/A 77.9% N/A

Combi. 86.8% 56.5% 102.3% 86.9%

Bilingual IR

Improvement … Query combination (concatenation) Relevance feedback

(yes for PT, FR, RU, no for FI)Data fusion (yes for FR, no for RU, FI, PT)

Multilingual EN->EN FI FR RU

1. Create a common index Document translation (DT)

2. Search on each language and merge the result lists (QT)

3. Mix QT and DT

4. No translation

Merging Problem

EN

<–– Merging<–– Merging

RUFRFI

Merging Problem

1 EN120 1.22 EN200 1.03 EN050 0.74 EN705 0.6…

1 FR043 0.82 FR120 0.753 FR055 0.654 …

1 RU050 1.62 RU005 1.13 RU120 0.94 …

Merging Strategies

Round-robin (baseline)

Raw-score merging

Normalize

Biased round-robin

Z-score

Logistic regression

Merging Problem

1 EN120 1.22 EN200 1.03 EN050 0.74 EN705 0.6…

1 FR043 0.82 FR120 0.753 FR055 0.654 …

1 RU050 1.62 RU005 1.13 RU120 0.94 …

1 RU… 1 RU… 2 EN…2 EN…3 RU…3 RU…4 ….4 ….

Merging Strategies

Round-robin (baseline)

Raw-score merging

Normalize

Biased round-robin

Z-score

Logistic regression

Test-Collection CLEF-2004

EN FI FR RU

size 154 MB 137 MB 244 MB 68 MB

doc 56,472 55,344 90,261 16,716

topic 42 45 49 34

rel./query 8.9 9.2 18.7 3.6

Merging Strategies

Round-robin (baseline)

Raw-score merging

Normalize

Biased round-robin

Z-score

Logistic regression

Z-Score Normalization

1 EN120 1.22 EN200 1.03 EN050 0.74 EN765 0.6… …

Compute the mean and standard deviation

New score = ((old score-) / ) +

1 EN120 7.02 EN200 5.03 EN050 2.04 EN765 1.0…

MLIR Evaluation

Condition A (meta-search)

Prosit/Okapi (+PRF, data fusion for

FR, EN)

Condition C (same SE)

Prosit only (+PRF)

Multilingual Evaluation

EN->EN FR FI RU Cond. A Cond. C

Round-robin 0.2386 0.2358

Raw-score 0.0642 0.3067

Norm 0.2899 0.2646

Biased RR 0.2639 0.2613

Z-score wt 0.2669 0.2867

Logistic 0.3090 0.3393

Collection Selection

General idea:1. Use the logistic regression to compute

a probability of relevance for each document;

2. Sum the document scores from the first 15 best ranked documents;

3. If this sum ≥ , consider all the list,otherwise select only the first m doc.

Multilingual Evaluation

EN->EN FR FI RU Cond. A Cond. C

Round-robin 0.2386 0.2358

Logistic reg. 0.3090 0.3393

Optimal select 0.3234 0.3558

Select (m=0) 0.2957 0.3405

Select (m=3) 0.2953 0.3378

Conclusion (monolingual)

Finnish is still a difficult language

Blind-query expansion: different patterns with different IR models

Data fusion (?)

Conclusion (bilingual)

Translation resources not available for all languages pairs

Improvement bycombining query translations yes, but can we have a better combination scheme?

Conclusion (multilingual)

Selection and merging are still hard problems

Overfitting is not always the best choice (see results with Condition C)

Multilingual Search using

Query Translation and

Collection Selection

Jacques Savoy, Pierre-Yves Berger

University of Neuchatel, Switzerland

www.unine.ch/info/clef/